From 17fd6e1843adfa11a9958fd21e056c304805bde2 Mon Sep 17 00:00:00 2001 From: YurenHao0426 Date: Mon, 10 Aug 2026 10:30:33 -0500 Subject: results: close feedforward PickupLoc gate --- results/babyai_shared/pickup_p0_analysis.json | 115 ++++++++++++++++++++++++++ 1 file changed, 115 insertions(+) create mode 100644 results/babyai_shared/pickup_p0_analysis.json (limited to 'results/babyai_shared/pickup_p0_analysis.json') diff --git a/results/babyai_shared/pickup_p0_analysis.json b/results/babyai_shared/pickup_p0_analysis.json new file mode 100644 index 0000000..5c21b79 --- /dev/null +++ b/results/babyai_shared/pickup_p0_analysis.json @@ -0,0 +1,115 @@ +{ + "best_clean_candidate_for_diagnosis": { + "bp_action_accuracy_percent": 84.81412488863691, + "bp_mission_lesion_drop_points": 52.0, + "bp_mission_lesion_success_percent": 11.600000000000001, + "bp_rollout_success_percent": 63.6, + "checks": { + "both_finite": true, + "bp_mission_lesion_drop_at_least_20": true, + "bp_success_at_least_70": false, + "clean_kp_success_at_least_60": false + }, + "clean_kp_action_accuracy_percent": 83.84222888150968, + "clean_kp_rollout_success_percent": 52.2, + "eligible": false, + "hidden_layers": 4, + "learning_rate": 0.03, + "source_files": [ + "results/babyai_shared/pickup_p0/d4_lr0.03_bp.json", + "results/babyai_shared/pickup_p0/d4_lr0.03_clean_kp.json" + ] + }, + "candidates": [ + { + "bp_action_accuracy_percent": 81.9875273345752, + "bp_mission_lesion_drop_points": 34.199999999999996, + "bp_mission_lesion_success_percent": 15.6, + "bp_rollout_success_percent": 49.8, + "checks": { + "both_finite": true, + "bp_mission_lesion_drop_at_least_20": true, + "bp_success_at_least_70": false, + "clean_kp_success_at_least_60": false + }, + "clean_kp_action_accuracy_percent": 80.30290758888799, + "clean_kp_rollout_success_percent": 45.6, + "eligible": false, + "hidden_layers": 2, + "learning_rate": 0.01, + "source_files": [ + "results/babyai_shared/pickup_p0/d2_lr0.01_bp.json", + "results/babyai_shared/pickup_p0/d2_lr0.01_clean_kp.json" + ] + }, + { + "bp_action_accuracy_percent": 79.03134364622986, + "bp_mission_lesion_drop_points": 31.4, + "bp_mission_lesion_success_percent": 13.0, + "bp_rollout_success_percent": 44.4, + "checks": { + "both_finite": true, + "bp_mission_lesion_drop_at_least_20": true, + "bp_success_at_least_70": false, + "clean_kp_success_at_least_60": false + }, + "clean_kp_action_accuracy_percent": 80.88604519316434, + "clean_kp_rollout_success_percent": 47.0, + "eligible": false, + "hidden_layers": 2, + "learning_rate": 0.03, + "source_files": [ + "results/babyai_shared/pickup_p0/d2_lr0.03_bp.json", + "results/babyai_shared/pickup_p0/d2_lr0.03_clean_kp.json" + ] + }, + { + "bp_action_accuracy_percent": 84.88701708917145, + "bp_mission_lesion_drop_points": 51.199999999999996, + "bp_mission_lesion_success_percent": 11.600000000000001, + "bp_rollout_success_percent": 62.8, + "checks": { + "both_finite": true, + "bp_mission_lesion_drop_at_least_20": true, + "bp_success_at_least_70": false, + "clean_kp_success_at_least_60": false + }, + "clean_kp_action_accuracy_percent": 81.9308334008261, + "clean_kp_rollout_success_percent": 51.0, + "eligible": false, + "hidden_layers": 4, + "learning_rate": 0.01, + "source_files": [ + "results/babyai_shared/pickup_p0/d4_lr0.01_bp.json", + "results/babyai_shared/pickup_p0/d4_lr0.01_clean_kp.json" + ] + }, + { + "bp_action_accuracy_percent": 84.81412488863691, + "bp_mission_lesion_drop_points": 52.0, + "bp_mission_lesion_success_percent": 11.600000000000001, + "bp_rollout_success_percent": 63.6, + "checks": { + "both_finite": true, + "bp_mission_lesion_drop_at_least_20": true, + "bp_success_at_least_70": false, + "clean_kp_success_at_least_60": false + }, + "clean_kp_action_accuracy_percent": 83.84222888150968, + "clean_kp_rollout_success_percent": 52.2, + "eligible": false, + "hidden_layers": 4, + "learning_rate": 0.03, + "source_files": [ + "results/babyai_shared/pickup_p0/d4_lr0.03_bp.json", + "results/babyai_shared/pickup_p0/d4_lr0.03_clean_kp.json" + ] + } + ], + "decision": "Add task memory and repeat a clean-only gate; do not run raw or SDIL on the present feedforward policy.", + "gate": "fail", + "raw_or_sdil_results_run_or_read": false, + "selected": null, + "stage": "babyai_pickup_p0_clean_feasibility", + "test_split_generated_or_read": false +} -- cgit v1.2.3